Citation: Liu, B.; Su, S.; Wei, J. The
Effect of Data Augmentation
Methods on Pedestrian Object
Detection. Electronics 2022, 11, 3185.
https://doi.org/10.3390/
electronics11193185
Academic Editor: Silvia Liberata Ullo
Received: 3 September 2022
Accepted: 27 September 2022
Published: 4 October 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
The Effect of Data Augmentation Methods on Pedestrian
Object Detection
Bokun Liu
1,
*, Shaojing Su
1
and Junyu Wei
1
College of Intelligent Science, National University of Defense, Changsha 410003, China
* Correspondence: liubokun16@nudt.edu.cn
Abstract:
Night landscapes are a key area of monitoring and security as information in pictures
caught on camera is not comprehensive. Data augmentation gives these limited datasets the most
value. Considering night driving and dangerous events, it is important to achieve the better detection
of people at night. This paper studies the impact of different data augmentation methods on target
detection. For the image data collected at night under limited conditions, three different types of
enhancement methods are used to verify whether they can promote pedestrian detection. This
paper mainly explores supervised and unsupervised data augmentation methods with certain im-
provements, including multi-sample augmentation, unsupervised Generative Adversarial Network
(GAN) augmentation and single-sample augmentation. It is concluded that the dataset obtained
by the heterogeneous multi-sample augmentation method can optimize the target detection model,
which can allow the mean average precision (mAP) of a night image to reach 0.76, and the improved
Residual Convolutional GAN network, the unsupervised training model, can generate new samples
with the same style, thus greatly expanding the dataset, so that the mean average precision reaches
0.854, and the single-sample enhancement of the deillumination can greatly improve the image clarity,
helping improve the precision value by 0.116.
Keywords: infrared and visible images; data augmentation; GAN; object detection
1. Introduction
Night pedestrian detection has great significance to the drivers, since, because of
the lack of light at night, driving vision is limited. It is difficult to distinguish pedestrian
positions. Moreover, night is the peak time for dangerous events, some intruders may
hide in the dark, and in most cases, situations caught by surveillance cameras are not
comprehensive, due to the limitations of visible light, camera jitter and rotation, making
it difficult to identify people in the dark. Visible image samples acquired under limited
conditions may have problems such as low definition and sample imbalance, and insuf-
ficient sample quality can lead to poor model robustness or insufficient generalization
ability. Therefore, to alleviate the above problems, data augmentation is a method worth
investigating. The essence of the data augmentation method is actually to make the existing
data more valuable based on the existing limited data, on the premise of not actually
collecting more data. When the sample data collected is not complete enough for objective
reasons, data augmentation methods can be used to generate data for new samples that are
more similar to the real data distribution, and elements such as noise or random images
can be introduced, so as to improve the recognition ability of the model and enhance its
generalization ability. At present, data augmentation methods are very numerous. In view
of images with poor lighting conditions, this paper studies the influence of different kinds
of augmentation methods on pedestrian detection and discusses the accuracy and objective
effect of images.
Data augmentation can be mainly divided into supervised and unsupervised data aug-
mentation. Supervised data augmentation, that is, enhance the data on the basis of existing
Electronics 2022, 11, 3185. https://doi.org/10.3390/electronics11193185 https://www.mdpi.com/journal/electronics
